def predict(img_bytes):
# Decode
x = image.decode_jpeg(img_bytes, channels=3)
print('decoded')
# Normalize
x = image.convert_image_dtype(x, float32)
print('normalized')
# Resize
np_x = image.resize(x, [224, 224]).numpy()
x = np_x.tolist()
print('resized')
# Predict
pred = model.predict(reshape(x, [-1, 224, 224, 3]))
# Map
top_k_values, top_k_indices = math.top_k(pred, k=3, sorted=True, name=None)
top_k_values = top_k_values.numpy().tolist()[0]
top_k_labels = [d[idx] for idx in top_k_indices.numpy()[0]]
results = {'prob': top_k_values, 'prob_labels': top_k_labels}
print(results)
# Return
return json.dumps(results)
def get_img(img_path):
img = read_file(img_path)
img = decode_jpeg(img, channels=3)
img = resize(img, [image_size, image_size])
img = 1 - img/255. # We would rather have the whole white void area be full of zeros than ones
img = tf.convert_to_tensor([img])
return img
def process_path(file_path):
print(file_path)
file = read_file(file_path)
file = image.decode_jpeg(file, channels=3)
file = cast(file, float32)
file = preprocess_input(file)
file = image.resize(file, [ROW, COL])
return file
def __call__(self, img):
if self.is_encoded:
img = tfimg.decode_jpeg(img)
img = tf.image.convert_image_dtype(img, tf.float32)
for cbr in self.comburant:
img = cbr(img)
if not tf.is_tensor(img):
img = tf.convert_to_tensor(img)
return img
def preprocess_image(self, image):
image = tf_img.decode_jpeg(image, channels=3)
image = tf_img.resize(image, [224, 224])
image /= 255.0 # normalize to [0,1] range
return image
def handle_image_path(img_path):
img = read_file(img_path)
img = decode_jpeg(img, channels=3)
img = resize_images(img, [image_size, image_size])
img = img / 255.0
return img
def load_image(image_path, img_width=299, img_height=299):
img = io.read_file(image_path)
img = image.decode_jpeg(img, channels=3)
img = image.resize(img, (img_width, img_height))
return img
def from_raw(cls, image_bytes: bytes):
image_tensor = decode_jpeg(image_bytes, channels=3)
return cls(image_tensor)